August 7, 2025. The Hong Kong Stock Exchange closed its session with a symmetry that should have alarmed more careful observers than it did. MINIMAX-W (00100.HK) ended the day nearly 25 percent higher. Zhipu (02513.HK) settled above a 17 percent gain. Two of China's most consequential large-model companies—recent alumni of the exchange's Chapter 18C listing channel, designed for pre-profit technology ventures—moved as though a single electrical impulse had passed through both tickers simultaneously.
There was no earnings surprise. There was no frontier-model release, no regulatory breakthrough, no enterprise megadeal disclosed in the intervening hours. The digits moved, and the feed stayed silent.
I have spent the better part of three decades reading financial infrastructure, first as a cybersecurity auditor inside a Sydney-based bank and later as an independent macro researcher tracking digital assets against global liquidity. I have learned that the market's most honest messages arrive in its silences. The silence between the digits holds the truth. And the truth embedded in this particular silence is not primarily about artificial intelligence—not about model architectures, benchmark scores, or training efficiency. It is about the liquidity that flows through the pipes connecting cryptocurrency speculation, Hong Kong's small-float listing regime, and the global re-pricing of Chinese technology assets.
We built castles on the tidal data of sentiment once before. In 2020, I watched Uniswap's total value locked surge past $2 billion and the assembled chorus of decentralized finance declare the arrival of a new financial order. I spent six months cross-correlating stablecoin issuance against global M2 money supply, and my data produced a verdict that rendered me unpopular in both camps: DeFi was not generating value. It was reflecting it—a polished mirror held to the face of fiat liquidity injections. The principle has not changed. It has only migrated.
Which brings me to the question this article intends to answer: what actually happened on August 7, when two Chinese AI companies with no shared catalysts, no interlocking business models, and fundamentally divergent technology roadmaps rose in unison—and what does it tell us about the architecture of the next market cycle?
The Two Companies and Their Divergent Grammars
MINIMAX enters the public market with a technical bet that is almost perverse in its audacity. While other Chinese model developers hedged their compute strategies across multiple vendors, MINIMAX committed to a deep, native relationship with domestic silicon—training its trillion-parameter mixture-of-experts model on Huawei Ascend 910B clusters. This was not a procurement decision. It was an infrastructure surrender to the reality of export controls, converted into a strategic advantage through sheer engineering persistence. The company had to build its own software toolchain from the kernel upward, adapting to the Ascend's immature software ecosystem and unstable cluster behavior. The result of that early, painful engineering is a rare asset in China's AI landscape: a frontier-scale model operationally proven on domestic hardware, with the cluster-stability scars to prove it.
The consumer surface of this bet is visible in Hailuo AI, a content generation tool that competes directly with Western generation suites, and Talkie, an AI emotional companionship product aimed squarely at international markets and the user base that Character.AI captured before its acquisition by Google. These are global products, monetized through consumer subscriptions and advertising, powered by a Chinese model stack with an unconventional hardware substrate. The route is one of consumer globalization, product velocity, and emotional engagement loops.
Zhipu speaks a different grammar entirely. Its lineage runs from Tsinghua University's laboratories, and the company has carried that academic inheritance deliberately into the enterprise. The GLM series long ago matured from research artifact to production infrastructure. GLM-4, in its various incarnations, has been deployed across government, financial, educational, and healthcare sectors—the high-friction, high-commitment institutional market that consumer AI companies rarely touch. Zhipu's commercialization strategy combines open-source weight releases for developer gravity with closed-source frontier APIs for revenue. It is the open-core play executed within the constraints of China's regulatory environment, where model alignment and content compliance are not afterthoughts but licensure conditions.
These two companies do not share a technology route, a customer profile, or a business model logic. They are not the same asset painted in different colors. The valuation grammar each requires is different. A consumer AI company should be priced on user growth, retention curves, and the psychology of digital companionship—metrics that resemble social media economics. An enterprise AI company should be priced on contract values, renewal rates, switching costs, and the depth of its institutional entrenchment—metrics that resemble defense software economics. The first is volatile and narrative-driven; the second is slower but stickier.
On August 7, the market treated these differences as noise. Both rose, in parallel, as though responding to a single instruction. And any analyst who cannot explain why two different assets moved identically should suspect, before assuming coincidence, that the market was not pricing the assets at all.
What a 25 Percent Single-Day Surge Actually Signifies in a Small-Float Regime
This is the point where my institutional training—the part that survived the Basel III disappointment—insists on structural specificity. A 25 percent daily appreciation for a widely held, heavily covered technology stock suggests a discrete event: an earnings beat, an acquisition premium, a product breakthrough. A 25 percent daily appreciation for a small-float Chapter 18C listing suggests something different: an order flow event.
The Hong Kong 18C regime permits listings by pre-profit technology companies with minimal public float requirements. This creates a visible feature of the Hong Kong AI market: the actual supply of tradable shares is minuscule relative to the narrative market capitalization. The mechanics are unforgiving. With a small free float, a modest absolute amount of buy-side pressure—a few institutional allocations, a coordinated southbound flow through Stock Connect, a hedge fund rotating into the narrative—can move prices by percentages that would require billions in volume elsewhere. The move is real, but its significance is ambiguous.
We have seen this pattern repeatedly in digital assets. The first weeks of any new token listing on major exchanges exhibited precisely this elasticity: low float, high narrative, double-digit percentage moves on order flow that would be trivial for any institutional investor. The market was not discovering a true clearing price. It was registering the presence of directional capital in a thin order book.
The crypto parallel matters more than most financial analysts would be comfortable admitting, because the information channel through which this stock surge reached the broader digital-asset community was itself a cryptocurrency platform. Bitget, an exchange-native outlet, carried the data to its user base. And this is the connective tissue the traditional narrative mischaracterizes: the marginal buyer of Chinese AI equities in 2025 is increasingly the same capital that has been rotating through digital assets. I advised the Reserve Bank of Australia in 2024 on the design of a digital Australian dollar, and during those closed-door conversations, the recurring motif was the blurring of boundaries—the way tokenized portfolios, stablecoin infrastructure, and conventional equity now share settlement rails and, more importantly, share sentiment vectors.
Liquidity is a ghost that haunts the ledger. It moves without announcing itself, showing up in one asset class as a function of its departure from another. The question for MINIMAX and Zhipu is whether the ghost has taken up residence or is merely passing through.
Reading the Surge Through the 2020 Lens
I cannot analyze the August 7 movement without summoning the analytical framework I developed during the DeFi Summer of 2020. The money printer was running at full capacity. Every asset class with a narrative attached was rising. And in the center of that storm, Uniswap's TVL crossing $2 billion was celebrated as validation of the automated market maker model—proof that decentralized protocols could aggregate liquidity without intermediaries.
My own analysis, published to a reception best described as muted, argued that the causal chain ran in the opposite direction. Stablecoin issuance was tracking global M2 expansion almost mechanically. The liquidity was not flowing into DeFi because of DeFi's merits; it was flowing because DeFi was a yield-obsessed expression of excess fiat supply. The TVL figures were not technology metrics. They were monetary indicators.
The same inversion applies, I suspect, to the MINIMAX and Zhipu surge. The trigger is not a discrete improvement in Chinese large-model capabilities that occurred in the previous 24 hours. The likely trigger is an allocation decision—generalist or cross-border funds rotating into Chinese AI exposure on the basis of a macro call: that Chinese assets are undervalued, that the AI trade has bifurcated into a US basket and a China basket, and that the China basket is early-cycle relative to its US counterpart. The companies, in this frame, are vehicles. The allocation is the story.
This is not inherently invalid. Markets routinely assign capital with coarse granularity, and the firms that benefit from a rising tide are not obligated to question the tide's origins. But for the analyst, the distinction between price appreciation driven by fundamental repricing and price appreciation driven by basket allocation is the difference between a durable trend and a tradable spike. The 2020 DeFi analogy instructs us that when the tide recedes—when M2 growth decelerated, when the institutional rotation paused—the assets with thin economic substance reverted with frightening speed.
In 2022, I watched $40 billion of value in TerraUSD vaporize in a single week, after months of algorithmic stability that its faithful had proclaimed the greatest innovation in monetary history. The underlying structure was sentiment. The outcome was structural. I do not mean to load the Terra analogy onto two legitimate companies with real technical assets. But the mechanism by which capital overprices a narrative—and then reverts—obeys the same physics at every scale.
What the Market Is Actually Pricing: The Basket and the Baseline
Let me be precise about what the market is pricing when it values MINIMAX and Zhipu through a shared lens.
First, the market is pricing China's AI policy credibility. The substantive alignment between China's industrial policy ambitions and the operational needs of large-model companies has produced a rare convergence: the state wants domestic AI champions, the companies want compute access and regulatory clarity, and the public markets are the mechanism through which this bargain receives its financial expression. Hong Kong's role as the offshore capital bridge for Chinese technology firms has been the subject of years of infrastructure building. The synchronous surge of two AI listings is a marker that this bridge is now open to the AI sector specifically.
Second, the market is pricing a winner's route to liquidity. The Chapter 18C listers have demonstrated that Chinese AI companies can exit to the public market. This matters for the entire early-stage ecosystem. For every MINIMAX or Zhipu that reaches the exchange, there are a dozen moonshot AI startups in Beijing and Shanghai watching these listings as templates for their own futures. The price level reached by the two companies establishes a valuation anchor, a floor of reference, for the next round of private fundraising across Chinese AI. The market's enthusiasm is thus not merely about two companies; it is about the revelation that the IPO window is open, and the entire sector's risk appetite adjusts accordingly.
Third, the market is pricing the comparative advantage of Chinese AI in consumer globalization. This is a MINIMAX-specific attribution, but its shadow falls on Zhipu through sector adjacency. The success of Chinese hardware in global markets—electric vehicles, drones, consumer electronics—has established a track record of Chinese companies executing go-to-market strategies abroad at scale. If Chinese AI products can replicate that pattern in software, and MINIMAX's Talkie, with its international user base and emotional-connection product loops, is the avatar of that ambition, then the valuations assigned to Western consumer AI companies become reference points for Chinese equivalents.
The set of these pricing dynamics is entirely plausible. It constitutes a rational foundation for the move. But the market has not confirmed that this is what it was doing on August 7, and the discipline of macro analysis demands that we differentiate between what a narrative can justify and what the order flow actually transacted.
The Contrarian Frame: Decoupling as Category Error
The conventional bullish reading of the August 7 surge treats it as evidence that Chinese AI has decoupled from its prior discount—that global capital has finally recognized the fundamental parity between China's frontier models and those of the United States, and is repricing accordingly.
I hold a different thesis. What we are observing is not decoupling from the discount; it is decoupling from fundamental anchoring altogether.
The phrase decorrelation has become popular in crypto and AI analysis, but it is often a category error. True decorrelation—genuine independence of price dynamics from the underlying asset's economic reality—is rare and generally temporary. The mechanism that typically produces a temporary decoupling is the substitution of one valuation narrative for another before the fundamental data arrives to adjudicate the difference. This is precisely what happened with the preceding framework for tokenized real-world assets. For three years, the blockchain industry attached itself to the RWA narrative, and the most deeply held institutional secret is that the traditional institutions never needed the public chain at all. The architecture was designed for the benefit of the distributed-ledger industry, not the balance sheets it pretended to serve. We told a story. Structures never followed. I see the same risk in the AI stock surge: the narrative is ahead of the underlying economic verification, and the structure that would justify the narrative—audited revenue growth, narrowing losses, a demonstrable product-market fit at scale—has not yet arrived.
The more specific risk embedded in the synchrony of the move is the conflation of differential technical routes. The market has placed MINIMAX and Zhipu in the same basket: Chinese AI core assets. But the two companies are not the same asset class. MINIMAX's valuation should be a function of consumer product traction, global user growth, churn, and the ability to monetize emotional engagement—a social-media-valuation frame with all its volatility. Zhipu's valuation should be a function of enterprise contract values, renewal rates, deployment dependency, and the replacement cost of entrenched incumbents—a public-sector-software frame with stickier but slower economics. These frames produce different revenue durability, different margin profiles, and different sensitivity to regulatory shifts. The market's synchronous treatment of the two suggests an index-fund mentality applied to companies that would be poorly suited to index-fund treatment.
And the deeper structural risk, which I cannot shake, lives in the open-source shadow. The archive remembers what the algorithm forgets.
Every frontier-model company currently operating confronts the open-source boundary. DeepSeek demonstrated in 2025 that a Chinese model could be released at dramatically lower inference costs, resetting the pricing power of the entire closed-source ecosystem. The economics of this are ruthless. If open-weight models converge on frontier capabilities, the closed-source premium collapses, and every model company's valuation is redenominated. This is not a hypothetical. It has already happened in the large-language-model sector once, and it will happen again. The market's memory, however, is short. It remembers the narrative, forgets the reversion. And on August 7, as the buying algorithm punched ticker symbols into the Hong Kong order book, I suspect none of the order flow was thinking about the open-source weight gradient.
Hong Kong as an AI Capital Venue: The Reinforcement Loop
The longer historical arc here deserves attention. Hong Kong was, for decades, a staging ground for the commercialization of Chinese industry. Its capital markets provided the mechanism through which state-linked enterprises, property developers, and hard-asset producers could access international capital. The subsequent restructuring of Chinese technology capital, with its pivot away from US listings and toward dual or Hong Kong-primary formats, transformed the city's exchange into the financial expression of China's national technology ambition.
The AI listings are the newest iteration of that transformation. But the venue imposes constraints that the participants do not always acknowledge. Hong Kong's public markets remain predominantly retail-driven compared with US institutional depth. The investor base is more regionally concentrated—Asian capital, much of it focused on China-AI exposure. The valuation markers are less internationally validated. An AI company listed in Hong Kong cannot automatically import the valuation grammar of Nvidia or Microsoft, because the investor base interpreting those grammar systems is not materially present in the same proportions.
This creates a feedback loop that is both a feature and a hazard. The resemblance of Hong Kong AI stocks to crypto-asset markets is not coincidental. Both operate within smaller, more narrative-driven investor pools, with higher volatility, lower float, and greater sensitivity to sentiment shock. The digital-asset lineage of the August 7 information flow—through crypto-native outlets to crypto-native audiences—reinforces this resemblance. If the marginal buyer of MINIMAX equity is a crypto-familiar allocating fund, then the buyer's risk model imports crypto volatility expectations, and those expectations, in turn, imprint on the stock's trading behavior.
There is a phrase I have used repeatedly with colleagues at the Reserve Bank of Australia while designing the digital Australian dollar: the transaction is cold; the trust is warm. It means that settlement mechanics are only half the infrastructure. The other half is the human confidence layer, which moves in sentiment waves. The August 7 surge was a wave. The infrastructure beneath it is still being built.
Opening the Shadow Ledger
Let me return to the event with the full weight of this analysis. MINIMAX and Zhipu moved in unison on August 7. I do not know the exact order flow, the composition of buyers, or the disposition of sellers. The data that would answer those questions—full broker-level tape, Stock Connect flows, margin-lending figures—is accessible but rarely analyzed in the immediate aftermath of a surge. I will therefore propose the questions that any serious analyst should apply to this event before accepting the bullish narrative.
Is the move backed by volume? A 25 percent gain on expanding volume, with high turnover and institutional-sized blocks, is meaningfully different from the same gain on a thin tape where a small number of funds determine the marginal price. Without volume data, the move's substance remains unverified.
Did the AI sector broadly move, or did these two names move alone? If the surge was sector-wide, with other Chinese AI names and related technology listings participating, the allocation thesis is stronger. If it was isolated to the two recent listers, the liquidity-mechanics thesis is stronger. I have not seen data confirming a broad AI-sector rally on that date, and the silence is itself informative.
What has changed in the fundamental trajectory of either company in the preceding 30 days? If nothing of consequence changed, the price movement is either early recognition of an existing thesis, which is plausible, or excessive order flow on stale information, which is dangerous. The distinction matters for follow-through.
Will the next earnings release confirm or contradict the market's new expectations? This is the only question that ultimately matters. A stock whose price rises 25 percent in a single day has effectively borrowed future returns. The market is betting that the companies will deliver results that justify the new price. If MINIMAX's quarterly user growth, retention, or revenue per user disappoints; if Zhipu's enterprise contract pipeline slows; if either company's cash burn accelerates beyond consensus—the reversion will be psychologically and mechanically swift.
We measured the shadow, mistaking it for the form. In our enthusiasm for the price movement, we mistook the shadow for the substance. The form—the companies themselves, with their technical platforms, their fragile ecosystems, their dependence on compute supply chains that remain geopolitically contested—has not changed in parallel with the stock price. The tension between shadow and form is where the risk lives.
The Cross-Contamination Hazard
Let me also be direct about the bearing of this story for the broader digital-asset and blockchain community, because the Bitget distribution channel connects it to a specific audience. In the bull-market climate of 2025, narratives travel faster because capital is abundant. The AI-stock surge in Hong Kong becomes, in crypto discourse, a signal of AI adoption by markets and often a justification for AI-token investments. I have watched these cross-contaminations before: a traditional-finance narrative informs a crypto-native restatement, and the crypto-native restatement—two layers removed from the underlying reality—functions as a distorted mirror of fundamentals.
The structural hazard is the belief that a rising price validates a technology thesis. It does not. A rising price validates liquidity, sentiment, and narrative circulation. The technology thesis is validated only by organic adoption, unit economics, and defensible advantage. The blockchain industry spent three years learning this lesson with RWA tokenization. It now risks repeating the same lesson with AI equities as the reference narrative.
I recall the moment I realized this with permanent clarity. In 2021, the NFT market exploded. Bored Ape floor prices passed $100,000. Creators celebrated a new renaissance of digital ownership and community building. I attempted to engage—sincerely—seeking the human connection and meaningful exchange the community claimed to represent. What I found was vanity and speculation: products designed not for use but for resale, communities organized not around mutual creation but around status competition. I withdrew for three months. When I returned, I wrote no more about consumer tokens. I redirected my research entirely into the infrastructure layer. The lesson from that period was a permanent one: sentiment is a weather system, not a climate, and no amount of positive emotion changes the intrinsic structure of an asset. The same weather system that produces a 25 percent surge can produce a 30 percent correction, and the only ground truth available to the analyst is the underlying value creation engine.
Between Knowing and Not Knowing
The August 7 surge has given the market a beautiful story. MINIMAX and Zhipu are real companies with real technology achievements. That the market has treated them as the avatar of a national AI ambition is not unreasonable. That the market has treated them as interchangeable is more questionable. That the move can be interpreted as decoupling from the underlying fundamentals is, in my reading, almost certainly incorrect.
Decoupling implies that these companies have achieved financial results that transcend the constraints of their environment—that they have become growth assets in the classic sense, their valuations driven by company-specific value creation rather than sector-wide liquidity. A 25 percent rise on no disclosure is not evidence of that. It is evidence of narrative momentum, of allocation flows, of the same ghost liquidity that has driven every asset market since the invention of the printed currency.
What happens next depends on the arrival of real data. The first earnings release from either company will be the moment of truth. The market's new expectation—created by the surge—will be tested against actual revenue, actual losses, actual user numbers. If the data confirms the re-rating, the surge transforms into a foundation. If the data disappoints, the surge becomes a spike, and the ghosts move elsewhere.
I have spent my career watching this pattern repeat—from the bank that could not see Bitcoin in its risk models, through the DeFi Summer that mistook liquidity for innovation, through the Terra collapse that mistook algorithm for stability, and through the NFT renaissance that mistook speculation for community. The machinery of reversion is invisible until it is fully deployed, and by then, the structural observer has already established a position elsewhere.
What I Am Watching Now
I am watching, first, the volume profile of MINIMAX and Zhipu in the five to ten trading sessions following August 7. A sustained high-volume advance tells me the order flow was institutional and committed. A fade on declining volume tells me the surge was a transient event.
I am watching, second, the southbound flow data. If mainland allocations through Stock Connect were the marginal buyer on August 7, the move carries a policy-adjacent significance. If international funds were the marginal buyer, the move speaks to a different phenomenon—global investors reassessing Chinese AI after a prolonged discount. These two explanations carry opposite implications for the next cycle of the trade.
I am watching, third, the next quantum of Chinese AI filings. If the 18C pipeline floods with new AI listings in the coming months, the surge was a sector-opening event, the establishment of a valuation baseline for the whole asset class. If the pipeline remains thin, the surge was a narrow event confined to two companies with limited consequences for the broader Chinese AI story.
I am watching, fourth, the earnings releases. Every company has a first earnings season as a public entity. For MINIMAX and Zhipu, those releases will transact more information than a hundred days of price action. The revenue curves, the gross margins, the cash-burn rates—these will write the next chapter of the Chinese AI capital story.
And I am watching, finally, the compute supply chain. Both companies are tethered to a chip-access reality that is fundamentally unstable. MINIMAX's Ascend dependency grants it a domestication advantage but also a single-supplier vulnerability. Zhipu's cost structure will be determined by its ability to acquire competitive compute from whatever mix of sources can be assembled under export control constraints. If the supply chain experiences a shock, no amount of market sentiment will matter.
Takeaway
The synchronicity of these moves is not a technical curiosity. It is a symptom. We are watching a market segment being repriced not on the basis of differentiated analysis but as a basket—and baskets are assembled for convenience, not for accuracy. As capital flees one narrative and migrates to another, the destination matters less than the fact of the migration. We have already seen the pattern in crypto assets, and we may now be seeing it in Chinese AI equities.
The question I leave you with is not whether MINIMAX and Zhipu deserve higher valuations. The more pressing question is whether the market that is assigning those valuations will still be present when the fundamental data arrives to verify the price. The transaction is cold; the trust is warm. And the temperature of trust, I have learned across three decades of watching the infrastructure gamble with the stories we tell ourselves, has a way of altering the shape of the ledger.